People love to quibble over cloud computing definitions, so let's lay this one out: This article's about infrastructure as a service, where companies rent virtual servers as needed for a variety of uses. Here, we're digging into gritty practicalities not often discussed, including pricing and contract rules.

Amazon.com's Elastic Compute Cloud service is the best known example, but the market now bristles with competition. To help companies understand this emerging market, we gathered detailed data from 12 infrastructure-as-a-service providers on prices, services, contracts, platforms supported, and more. A complete report of these findings, including four pages of comparison data on the offerings, can be downloaded free.

We can draw some conclusions from that data and from what we're hearing from customers. Hosting Web sites and testing software are the most common uses today, but public cloud services will be used more and more for variable workloads such as research or for peak e-commerce demands.

Almost all providers use either the VMware or Xen hypervisor to spin up server instances--only one of the 12 we researched supports Microsoft's Virtual Server, though that will change as Microsoft extends its influence in this market with Hyper-V. The providers support a variety of operating systems, Web servers, and databases, and they can slice physical resources into discrete chunks that can be allocated to one or many customers. All 12 providers let companies buy very small increments of computer power (as little as 256 MB of RAM) and pay only for the memory, disk space, and network bandwidth they consume. Only three service providers on our list, AT&T, IBM, and Unisys, require an annual contract.

With infrastructure as a service, customers pay for computing power based on consumption, much like electricity. "We think of cloud computing not as a new technology, but as a new purchasing paradigm," says Savvis CTO Bryan Doerr. It's also the place to expect cutthroat competition. Prices today run as low as 1.5 cents per hour for a CPU from Rackspace. While there's plenty of room for providers to differentiate on type of service, and thus charge premiums, expect price competition to remain intense given the number of market players.

Savvis, for example, jumped into the market earlier this year by adding virtualized server instances to its mix of hosting services. Verizon launched a cloud infrastructure service in June, and Unisys a month later. Rackspace has established itself as a cloud infrastructure force, and smaller companies such as 3Tera and Skytap are selling more specialized services by the computing slice.

The Appeal Of On-Demand CPUs
Web hosting and testing and development are the two most popular uses for infrastructure as a service. If an e-commerce site has its Web servers in a cloud and traffic loads increase, more Web server resources are available on demand, so it doesn't have to buy capacity only for peak demand. When the burst ends, those Web servers are eliminated, along with the hourly rate to run them.

Mike Casullo, CIO of satellite broadband provider WildBlue, turned to the cloud because the company didn't have enough test slots to work on all the projects coming out of development. "It's like you're driving around a parking lot waiting for a space to become available," he says.

Casullo didn't want the hardware or maintenance expense of expanding WildBlue's testing and development environment. So he supplemented it with Skytap, which lets customers create a virtual lab by uploading their own images or using pre-configured images of an OS, virtual machine, and application. The payoff: In four months, WildBlue spent only $9,500 on Skytap for a testing project, instead of more than $500,000, Casullo estimates, if it had to buy, provision, and administer hardware itself.